Wind power expansion in the Brazilian Northeast faces two mutually reinforcing difficulties: the best-resource areas are progressively being occupied by operating projects, and the grid-connection cost has become comparable to the cost of the turbines themselves. Treating site selection and layout optimization as separate problems, which is the dominant practice in the literature, leads to suboptimal solutions under these conditions. This paper proposes Wind Farm Swarm Optimization (WFSO), a bi-level formulation based on Particle Swarm Optimization that unifies both problems in a single search loop: the outer level moves the centroids of candidate farms in continuous space, while the inner level optimizes the turbine layout of each candidate. Both levels share the levelized cost of energy (LCOE), which endogenously incorporates the connection cost, automatically selecting from among the available scenarios, and the interference zones of existing wind turbines. The methodology was validated on a benchmark case with a known optimum and applied to a 3500 km² region in Rio Grande do Norte, Brazil, using real resource, constraint, and infrastructure data. Four outer-PSO settings, spanning 1300 to 20,200 objective-function evaluations, converged to the same region and connection point, with a reference LCOE of 249.64 R$/MWh under a common seed. Repeated runs with independent seeds reveal a bimodal objective landscape whose success rate is governed by the evaluation budget: from about 11,400 evaluations, every seed reaches the best basin, and a configuration that is executable in 3.6 h stays within 0.27% of the 38.6 h reference. A sequential siting-then-layout baseline built with the same inner optimizer reaches an LCOE 9.4% higher, and on a reduced benchmark the method matches the optimum verified by exhaustive enumeration in 10 out of 10 seeds, which makes the tool practical for prospecting campaigns covering multiple areas.

Wind power expansion in the Brazilian Northeast faces two mutually reinforcing difficulties: the best-resource areas are progressively being occupied by operating projects, and the grid-connection cost has become comparable to the cost of the turbines themselves. Treating site selection and layout optimization as separate problems, the dominant practice in the literature, leads to suboptimal solutions under these conditions. This paper proposes Wind Farm Swarm Optimization (WFSO), a bi-level formulation based on Particle Swarm Optimization that unifies both problems in a single search loop: the outer level moves the centroids of candidate farms in continuous space, while the inner level optimizes the turbine layout of each candidate. Both levels share the levelized cost of energy (LCOE), which endogenously incorporates the connection cost, automatically selected from among the available scenarios, and the interference zones of existing wind turbines. The methodology was validated on a benchmark case with a known optimum and applied to a 3,500 km² region in Rio Grande do Norte, Brazil, using real resource, constraint, and infrastructure data. Four outer-PSO settings, spanning 1,300 to 20,200 objective-function evaluations, converged to the same region and connection point, with a best LCOE of 249.67 R$/MWh; a configuration executable in 18 h stays within 0.39% of the 11.3-day reference, which makes the tool practical for prospecting campaigns covering multiple areas.